Cultural context and the case for customized LLMs

A user argues that current LLMs start each session with no cultural context, making it expensive in tokens to get them into the right frame of mind. They predict this burden will be as much a driver of model customization as data security. In a follow-up, they claim productive reasoning about complex strategy is essentially impossible without loading more than 400,000 tokens and repeatedly challenging models to use the document collection rather than fill conceptual gaps from their training.